TW491938B - Method of designing a multichannel filtering system for measuring digital camera's spectral responsivities - Google Patents

Abstract

For determination of camera's spectral responses, we introduce a multichannel filtering (MCF) system. The design objective of the optical system is to effectively select a limited amount of spectral (or broadband) filters to characterize the spectral features of color imaging processes, which are contaminated with noise, so that the spectral response functions can be estimated with satisfactory accuracy. In our approach, a theoretical study is first presented to pave the way for this work, and then we propose a filter selection algorithm based on the technique of orthogonal-triangular (QR) decomposition with column pivoting (QRCP), called QRCP-based method. This method involves QR computations and a column permutation process, which determines a permutation matrix to conduct the subset (or filter) selection. Experimental results reveal that the proposed technique is truly consistent with the theoretical study on filter selections. It is found that the MCF system with the filters selected from this method is much less sensitive to noise than those with other spectral filters from different selections. Thus, the spectral responsivity estimation can achieve a satisfactory accuracy.

Description

491938 V. Description of the invention (l) 1 · Background of the invention: (1) The scope of the invention: The present invention relates to the design method of a spectral response measurement system, especially for digital image pickup devices (for example, digital cameras, cameras, etc.) A method for designing a spectral response measurement system. More specifically, the present invention is a method for designing a multi-channel filtering system for measuring the spectral response of a digital imaging device. (II) Prior technology: The color signal quality of general digital image pickup devices is mainly determined by the spectral response defined in the visible light range (400-700 nm). This spectral response is related to its optical elements, spectral filters (for example, infrared cutoff filter), color filter array (color filter array), and image sensors (imager). The spectral characteristics are related. If the spectral response of the color imaging device is known, we can use it to develop important and economical technologies and applications of β Midnight. For example, computer (mechanical) vision, color measurement and reproduction technology, video processing, pattern recognition, and color image processing technology, etc. [丨] — [9]: More specifically, we can calculate the color of this device The quality factor (c0rimetric qualify factor). This color quality factor was established by Neugebauer [9] .4 It is used to represent the approximate process I of the spectral plasticity of mechanical vision and human visual system. Determining digital cameras or camcorders :: The quality f of the image to be captured is very important for the quality control and analysis of digital image capturing devices

49193.8 V. Benefits of Invention Description (2). Therefore, measuring the spectral response of a color imaging device is of great industrial application value. Unfortunately, usually the overall spectral response of a color imaging device is difficult to obtain (even though we can measure the spectral response of an image sensor). The results from various investigations show [8] that the main reasons for the slow development of the spectral response measurement technology of these vision systems can be summarized as follows: (1) the presence of camera noise; (2) the lack of manufacturable and inexpensive color measurements Colori-metric filter; (3) manufacturing limitations and difficulties of optical (or color) filters (for example, the effect of differences in filter thickness on transmittance; similar to semiconductor processes, this spectral response is affected by color The effect of variations in the process parameters of the filter array is significant; therefore, it is difficult to determine whether the spectral response of the color channel meets the required specifications). (4) Imperfect (or not ideal) spectral distribution of optical components in nature; (5) Due to the nature of multi-channel color imaging (muitichannel), the complexity of the integration of optoelectronic systems is caused. Due to the existence of these factors, the color accuracy of the image pickup device is quite limited, even if the relevant color standards have been established internationally; for example, the specifications recommended by the International Lighting Association (CIE) in colorimetry ' And the National Television System Committee (NTSC) color standards [16].

Abraham and Wenzel [3] in their US patent proposed a technique for measuring the spectral response parameters of human eye color perception. Takaha-

Page 7 491938 V. Description of the invention (3) Shi and Terashita [4] invented a method to estimate the spectral response and exposure of traditional camera films. Osaki and Sugiyama [5] published a colorimeter spectral response correction device, which adjusts the output value of the colorimeter to be measured based on the assumed spectral response, but does not actually measure the colorimeter's spectral response. In addition, Zerlaut et al. [6] and the International Illumination Society (cie) [13] separately proposed several spectral response measurement techniques for optical radiation detectors. Unfortunately, these methods are difficult to directly measure the spectral response of digital color imaging devices. 〇 ^ et al. [1〇] used interferometric filters (1111: 61 ^ 61 ^ 1 ^ 6 ^ 1 to 1 ^) or narrow-band filters to measure this spectral change. This filter has a characteristic of separating a narrow wavelength passband (narrow wavele ^ gsth band), and can be used to measure the spectral response of an imaging device. To meet this characteristic, this method requires the construction of a more complex and sophisticated optical system. ^ In addition, these filters are also much more expensive than broadband (or absorption) filters.

Once traditionally, we can use a single light meter (m〇n〇chr〇mat〇r) to adjust the spectral response of the photoelectric sensor (ph〇t〇-electric detect〇r). The experimental mechanism of the optical instrument or dry-type filter 'set up 1 H, wyszecki [12] took the lead in proposing a wideband pass filter (br0ad_ 2 1k, # Λ 7Γΐη§) method. In his method, 14 spectra The filter sees the conditions in the range of light formed by its transmittance distribution-group line 2813. However, the International Illumination Association (c IE) [1 3] ·;: under the two St limiting factors (eg, miscellaneous Information exists), the filter response is used to measure the spectral response of the sensor, and its accuracy is acceptable;

491938 V. Description of the invention (4)-will be reduced (because the spectral transmittances of these filters in the visible range overlap each other quite closely). In addition, narrower absorption band-pass filtering has been replaced by narrow-band interference filters (the narrower spectral passband formed by the series combination of absorption filters will increase with the overall thickness, which is prone to failure. Transmission rate) [21]. Therefore, in order to overcome the above problems, we constructed a set of spectral response measurement systems based on multi-channel broadband filters [7], [丨 4]. In the present invention, we further propose a clever filter selection method to efficiently design this filter-based optical system to reduce the effect of noise and thereby improve the accuracy of the spectral response measurement. [丨 5]. 2. Summary of the invention: (1) Purpose of the invention: The purpose of the present invention is to provide a method for designing a spectral response measurement system for a digital imaging device to establish a set of spectral response quantities with high accuracy, simple system structure and low cost.测 系统。 Test system. ~ Another object of the present invention is to design (or select) an optical system (called a multichannel filtering (MCF)) system based on a limited number of #broadband (broadband) spectral filter sets. Establish a spectral response measurement system to reduce the sensitivity of this optical system to camera noise, and then effectively estimate the spectrum of digital imaging devices

Page 9

49193S V. Description of the invention (5) Response. (II) Brief Introduction of the Invention: In order to achieve the purpose of the present invention, we first briefly describe the use of a multi-channel filtering (MCF) system to perform a spectral-response characterization process on a digital imaging device. The results of the sensitivity analysis of this system for noise are then cited to illustrate the effect of camera noise on the accuracy of the spectral response measurement. In addition, we use matrix calculation theory to illustrate the feasibility of using a spectral filter to reduce the sensitivity of the system to noise, so as to perform an efficient system. From this theory, it is known that the selection of filter sets in system design can be used to improve the accuracy of spectral response estimation (that is, reduce the system's sensitivity to noise). In the present invention, we propose a filter selection calculus based on orthogonal-triangular (QR) decomposition with (CDlunrn pivoting (QRCp)) technology, which is called QRCP-based method. This method involves a process of orthogonal triangle decomposition and row permutation transformation (co! Umn permutation), where this process determines a column moment, which is used to dominate the selection of this subset (or filter). The results of the Bayesian test show that this technique is consistent with the filter remote selection theory discussed. As we expected, the optical system constructed by the filter selected by this QRCP-based method has far better noise tolerance than that constructed by other filter selection methods, making this hidden system have a satisfactory spectral response. Estimate accuracy. 491938 V. Description of the invention (6) 3 · Technical contents: (1) Details: (1) Multi-channel filtering (MCF) system · Figure 1 shows that we have come to the proposed multi-channel filtering (MCF) system. Characterization of the spectral response of a digital camera. This process is used to generate a collection of incident light (called color stimulus) with color to excite the camera so that its color shirt features can be truly extracted from the output of the camera. In this process, we capture the spatial average of the light radiation volume of a color stimulus in a limited area, so the color stimulus can be regarded as a function of the wavelength (body "161 ^ 1 ^). This color stimulus set is labeled ), ί = 1, 2, ..., Qin} is used to indicate its spectral power distribution (SPD), the unit is W sr'1 m'2 nm4. Results from various experiments and investigations [8 ], [14], we made the following reasonable assumptions for the camera under test in this system: (a) The degree of spatial distribution of the camera's spectral response can be ignored; (b) No blooming occurs; (c ) The relationship between the camera output and the incident light signal is sufficiently linear (for example, you can cancel the camera gamffla correction function to avoid

Page 11 491938

Free from non-linearity). In addition, all camera parameters (such as aperture size, white balance setting, etc.) are fixed. In the process of characterizing the spectral response, relative to the first color stimulus, we represent the output of the camera ’s a channel as [8] heart Γς (jealous W1 +, + jealous U: 1, 2, 3, 1 Mathematical formula (1) where '⑷ is the camera's spectral response, which is the offset caused by the dark current, # is the camera noise [17], and the visible wavelength range [λ, λ2] = [40 0 nm, 70 0 nm], and channel α: = ι, 2,3 are usually expressed as three channels of RGB (red-green-blue). In the kMCF system, we use a spectral filter (Spectrai fi 1 ters) (W ) J = i, 2, ..., Nc to modulate the radiant light emitted by a diffused area light source ^) to generate this color stimulus set. Figure 2 shows a preferred implementation example of the spectral power distribution of this light source [7]. Consider that the light emitting surface of the light source and the surface of the filter are geometrically parallel to each other. Therefore, as shown in FIG. 1, the generation of the color stimulus q 可以 can be expressed as G (again) = 5 (2) / (2), έ = ι’2, ... Mathematical formula (2) Obviously 'formula (1) can be expressed in the form of a vector (or matrix),? I = \ X ... rNc? I = 1,2,3, mathematical formula (3) where & = An is called the trimmed obser-

Page 12 491938 V. Description of the invention (8) vation), # and A are the column vectors representing ~, and the dimensions are in the visible spectral range with wavelength resolution (quadrature) sampling), and the superscript Γ represents the transposition of the vector or matrix. In order to solve the spectral response vector #, the value of ~ must be greater than or equal to the heart value, and considering the convenience of calculation, we set I = Further, the formula (3) can be written as «== (: # + /, BU3, mathematical formula (4) where the vector < occupies the position of the i-th column of c when the color stimulation moment is occupied, that is C 2 仏] , And e ^ xl is the channel-first noise vector. We quickly observe that the vector e, which is the spectral response, is poor and can be estimated 'if rank (C) is satisfied, where rank (_) is the matrix's Matrix rank. (2) SyStem sensitivity to η oise analysis. Singular value decomposition (SVD) plays a very important role in searching the matrix structure. Consider For the matrix e Orthogonal matrix // ear movement and f eir- make mathematical formula (5) ^ TAV = diag (σι »σ2—) e

Page 13 491938 V. Explanation of the invention (9) = number, = min (w'0 'min (·) means taking the minimum value, diag (·) represents a diagonal matrix, and the elements on its diagonal are called singular Value. We label the moment ⑽: the large singular value as ㈣. In the expression ⑸, F is a matrix formed by the right singularity ht s ^ ^ ular vect0r), we call it the right singular moment by singularity As a result of the value decomposition, we introduce the definition of the condition number to obtain an accurate measure of noise sensitivity. Let the square matrix (clu) be non-singular, and mrank (c) =%, then the condition number of c is marked as, where -2 (C) M 1 ^) 1 = σι (〇 Κ (〇 Mathematical formula (6) Mathematical formula (7) 5 (c) Tune 2 and% (c) = i / || (c) l and 丨 2 «2 represents the modulus value of the matrix (or vector) (2_n 〇rm) [18]. If c) a 1 is relatively small, it is commonly known as well-contained.

In this MCF system, in order to reduce the effect of noise, it seems that we can repeatedly observe the rotation of the camera to obtain more information, and then use ordinary digital estimation to calculate the spectral response function (or vector). However, this approach is impractical because more observations can cause the characterization of the spectral response to be more time consuming. It is known from equation (4) that by matrix calculation

Page 14 V. Description of the invention (10) Theory We can prove that if the condition number h (c) of the matrix C is sufficiently small, the estimation of the spectral response vector can have quite satisfactory accuracy. This result is Because this system is as insensitive to noise as possible [i 5], [18, theorem 2 · 7 · 2 and Lemma 2.7 1]. From equations (2) and (4), we find that under the known light source, the spectral filter corresponding to the color stimulus determines the structure of this Mcf system and the sensitivity of the system to noise. Specifically, proper selection of the filter can make the minimum eigenvalue% (C) of the matrix C sufficiently large, and according to this result MO = a1 (C) AMC) will become as small as possible, where = || C || 2 is related to the total power of the color stimulus set. Due to the broadband nature of the color stimulus, this value is limited to a relatively limited range (ie bounded). Therefore, the choice of the 'filter is very important for the accuracy of the spectral response estimation and the sensitivity of the system to noise. (3) Design of efficient MCF system (a) Theoretical discussion of spectral filter selection From the above, it can be known that the 'wide-band color stimulus signal' can be represented by a histogram of a vector with a wavelength resolution of ~ = heart. In general, a huge set of color stimuli can be represented by a matrix j € sr " with row partitioning, where a a 乂 is the number of these color stimuli and the wavelength resolution. It is clear , Each column vector of the moment 代表 represents an individual color stimulus. 1

Page 15 49193.8 V. Description of the invention (11) In this matrix 2, a submatrix formed by arbitrarily selecting a row vector can be regarded as the realization of a set of color stimuli or regarded as a kind of spectral filtering corresponding to color stimuli Implementation of the device selection result. For subset selection, we briefly describe an important result involving permutation matrix in matrix calculation theory [18]. Considering the singular value decomposition (SVD) of j e obtained by equation (5), we have the right singular matrix F of j: ^ -Nc where rankU) is lacking. Define ping pong, with the formula (8) ι-Νζ

Where / > is FV

Vn VnΛ 4 Nc n ^ Nc Mathematical formula (9) Permutation matrix. Let η-Μ 〃 mathematical formula (10) and if κ is nonsingular, we have σ ^ (Α) / ψ \\ I < ^ σ ^ (, ίΐ). From the equations (8) to (li), we know that the permutation matrix ρ plays a very important role in the selection of the subcollector. This matrix records the action of column interchange of the second string. For example, the second row and the third row of the matrix are to be interchanged at the same time. I come to M, and it is formed by the two rows interchanged with 4. Moment $ 王 卞 "can be expressed

V. Description of the invention (12) Shown as a name, that is, a row transformation expression of a moment ❼ can be expressed as a permutation matrix C ^ π " 。By single [Tow song 枘 ΥΥ matrix) 'e # " line 2 and line 3

Reciprocally, we can get the contents of this matrix ^. , —VI also corresponds to a series of permutation matrices P | p v, which can be represented by a permutation matrix 1 Qi ... In the expression (9), the matrix / > 番 报 戈 庠 # ^ ^, 'the submatrix of 1 and 4 in the rows r / 排 of the matrix 4 (column) appears in front of # A block formed by a single column and the remaining 6W r ® "column" is marked as A. For formula (4), you can move AW __ b · a large value to the ground. I ^ c = Years, and assuming that π) has coherent expressions ⑹, (10), and ⑴), the zhen learned that the decision to arrange matrix two can make IMU ice) as small as possible, that is, ⑽) is sufficiently interesting. This-result through the operation of singular value decomposition (svd), we can know that =, (〇 = training will become large enough, that is, ⑽ will become small enough. Therefore, the decision matrix * is a wavelet (or a subset) (B) The algorithm of filter selection is designed in the matrix calculation theory [1 8]. We found that the qRCP method is an effective method to determine the matrix P, so that 阼 I in OR (11) becomes Small enough. Then based on this method, we developed the QRCp_based method to select 491938. V. Description of the invention (13) Choose the finite number of spectral filters you want It is used to construct this MCF system to reduce the system's sensitivity to noise and improve the accuracy of spectral response measurement. (I) QRCP method: Consider the orthogonal triangle (QR) of a matrix decomposed into ^ G = duty or QrG Mathematical formula (12) makes the matrix 0 旷 is orthogonal ㈠ "dagger ⑽" and the matrix "mine η is upper-triangular, where" and "are integers. This "method" includes QR decomposition and row permutation (column pivoting or column permutati). For row conversion, a series of row swapping actions will result in rows with smaller non-square values (2-nonn). (Column) moves to the right corner of this matrix. In addition, for QR decomposition and ancient times, a series of steps of row permutation conversion are combined with reflection operator (transfc) rmatic n)). For a detailed description of the operation of QRCP, please refer to the reference [丨 8] 〇From formula (8), let us

Vr- Mathematical formula (13) Mathematical formula (14), upper triangle moment Operate QRCP method on γ, get er [K] P =

Nz n, /, middle-right parent diagonal matrix ρ, permutation matrix ^ 491938 V. Description of the invention (14) Matrix A ㈣, and meaning is a meaning X and W matrix. For sub-set (or filter) selection, the matrix heart plays an important role. 1. We label the elements on the main diagonal as slices, hi, 2 ,,%, where Fan 0 is non-singular because of the two-sided matrix!?!. Due to the column interchange effect caused by the matrix / ^, the absolute value of the slice, that is, kl, hU, .., 乂, will become relatively large. The row vector (c lumn vector) is obviously independent of each other. Through SVD calculation, it is found that the ratio of ^ &) to 〇, #,), that is, & condition), is sufficiently small. Therefore, the process of permutation and transformation of this line will make the meaning as well-conditioned as possible. From the formulas (10) and (14), for 矻 we have ίκ, Ί = ρτ X " k. 'L from 0 mathematical formula ( 1 5) Due to the nature of the direct parent (〇rth 〇g ο na 1) 6r0, we can prove the mathematical formula (16) [18, ρ · 574]. From formulas (6), (7), and (16), we can get the light 'L to take _1. Mathematical formula (i 7) because ^ is as much as possible ^^ 11-〇011 (^ 1 ^ 0116 (1, And the ratio of # 〇 to < ^ 1 |) is bounded (b 0unded), so it will become quite 11-conditioned; that is, 6 under, or IL is small enough. So this

Page 19 491938 V. Description of the invention (15)-The QRCP method is an effective method for selecting a subset (or filter). (i i) QRCP-based method: We summarize the effect of the above QRCp method on the selection of subsets (or filters) as five inductive points, as shown in Figure 3. According to the causal relationship, these induction points can be arranged as follows: • Induction 1: QRCP operation that cannot be taken; Induction 2 • makes% 洱,) sufficiently small, where A is the upper triangular matrix; induction 3: leads to 5%) and IPl is sufficiently reduced (as known from Equation (7)); Induction 4: makes the value sufficiently large; Induction 5 • As can be seen from Equation (11), ~ (成) will be close to 0 ~ (『). If% ⑼ is large enough, = can be sufficiently increased. Therefore, as shown in equation (4), the spectral filtering selected by this method as the iMCF system will be as insensitive to noise as possible. In the present invention, we propose a QRCP-based filter selection algorithm, called the QRCP-based method. As mentioned earlier, the matrix J € is formed by the spectral power distribution (SPD) of "color" stimuli with wavelength resolution ^, where H = #. To identify these # candidate filters, we will separate them Mark the filter index number (filter index number) 1,2, ···, #, and let the wave wave be, and the cable bow 丨 inward (filter index vector) = [12 € s ^ h | A collection of numerical order. What we mentioned

491938 V. Description of the invention (16) The QRCP-based filter selection algorithm is shown in Figure 4. The filter selection algorithm: Yu Qinsu is as follows. Step 1: The SVD operation of matrix 4 and stores the matrix p, as in the formula (5) Step 2: Check whether rank (jlw is satisfied. If yes, go to the next step. If it is not (N0), end the work. Step 3: Take the rows in the K matrix (c 〇lumn), to construct the sub-moment bass, this K is called the right singular sub-matrix. Step 4: Take the QRCP operation of C, and store the permutation moment 呻 p. Step L Lets < The vector formed by the trans-posed filter index vector is arranged in rows. Specify the vector d, followed by a $: to form the vector f. These elements are the selected filter index number. therefore. , ≪ is called the selected filter index vector (selected index vector). Γ followed by the work (e n d).

(B) Preferred implementation example: We use # = 135 Roscolux spectral filters to construct this huge color stimulation matrix 2. Using this QRCP-based algorithm, we get

Page 21 491938

H4 selected wavelet [15], its spectral transmittance distribution is shown in the figure. The index numbers of these selected ferrules are shown in bold line 2. 4 Yi (1) The estimated spectral response function (or vector) of the spectral response vector (element U3) is actually nonnegative, that is, this limit can be expressed as, ^ ^ > 0 A : = L, 2,3 〇, 'Mathematical formula (18) is derived from formula (4), in order to obtain the best solution of the spectral response vector estimation from the viewpoint of the least square error, mark {i (* U = U3 }, We use the NNLS (n0nnegative least squares) algorithm proposed by Lawson and Hanson [19] to solve the following NNLS problem: minimize (lCsiA), mathematical formula (19) satisfies evening) 20, where the spectrum The response estimation vector ⑼U = US} is a vector of 乂 xi.

Based on the estimated spectral response vector w, we can use perfect reconstruction or upsampling techniques [7], [20], ie, low-pass filtering (i0w) ~ Pass filtering) technology to get a spectral response with the desired wavelength resolution. Figure 6 shows the estimated spectral response of the Sony XC711 camera (wavelength resolution of 5 nm).

Page 22 491938 V. Description of the invention (18) (2) Examination of noise effect: Taking the estimation and results of different filter selection methods as examples

In order to verify that this QRCP-based method is superior to other selection methods, we consider three other different filter bank selection results, which are labeled as selection_1, selection2, and selection3. (se 1 ec ti on 3). Their filter index numbers are shown in lines 3, 4, and 5 of Table 1, respectively. In this table, the index numbers of the filters selected by the QRCP-based method are shown in bold bold. Selection method 1 replaces one of the filter sets selected by the QRCP-based method with a different filter (for example, index number 15). The number of this different filter ^ 丨 We underline the number to indicate that this filter is different from the one selected by the QRCP — based method. In the same way, the filter selected by selection method 2 is half different from the one selected by this method, and the selection method 3 is completely different from this method. Table 2 shows that our experimental results are consistent with the above-mentioned filter selection theory, that is, satisfying the relationship of equations (11) s £ and ⑼. Table 3 compares condition " numbers from different filter selections. Obviously, the condition number obtained by QRCP-based method is the smallest, which shows that the KF system constructed by the filter set selected by this method has the least spectral noise estimation result. In order to test the effect of noise under different signal-to-noise ratios (signal_t〇_n〇ise rati〇 (SNR)), we use the spectral response estimated in Figure 6 to simulate the camera's spectral response (simuUted spectrai

491938

responsivi ties), and considering the use of approximately 8-bit image capture systems, the signal-to-noise ratios that the system may have are 30, 35, 40, and 4 5 dB [22]. Table 4 shows the normalized root-mean-square error (〇veraH n〇rmaiized root-mean-square error) of the MCF system constructed by the filter set selected by the QRCP-based method under different SNR ratios. NRMSE)) are all smaller than the results of other wave filter choices. Taking SNR 35 dB as an example, as shown in Fig. 8 (d) and 8 (a), 8 (b), 8 (c), the estimated and simulated spectra obtained by the qrcp-based method are compared. The response has the smallest error (compared to the results of spectral response estimation of MCF systems constructed by other selection methods 1, 2, and 3). Therefore, we can see that the theoretical, simulation, and experimental results of this method are better than other filter selection methods.

(3) Verification of spectral response estimation performance (or error) In order to calculate the spectral response estimation error, we propose a color measurement subtraction based on the 0/45 color measurement architecture [21] proposed by the International Lighting Association (CIE). With 1 degree (〇: 〇〇〇): testing and measurement (C T M)) group to generate 7 2 different test color light (seven es seven ing colored 1 ight (TCL)). This test color light collection consists of 3 different light sources (tungsten halogen lamp (THL)) with a color temperature of 3 200 K, daylight-type illuminators D55, and D65, to complement the Macbeth Color Checker (MCC). Of 24 standard color films. Figure 7 shows the normalized root mean square error (NRMSE) distribution with respect to test color light (TCL) numbers 1 to 72. Table 5 lists under this CTM configuration.

= The definition of the normalized root-mean-square error (Qverall NRMSE) of the spectral response estimation error obtained by the wave filter selection is b 'in the table,' orth | | Σ 2 (Ww γ / W3 = l. Where is the estimated camera output value帛 Under the formula (20), the spectral power of the light is measured by a known color power sheep knife: and the estimated spectral response is matched with the formula. In Table V: to) / "is the actual The camera output value and ~ = 72 = person ^ construction saw that the filter selected by the fan method ㈣ 均 mean square I】 obtained by the maximum phase value of all color channels in the spectral response gauge (MSE) are less than 0.022, And in these channels, the normalized estimation error (maximum normalized estimation err) is taken to be 0.000425. 4 · Efficacy and Features: The comparison between the present invention and the prior art is as follows:

(, ≫ a) We propose a set of spectral response measurement system (ie MCF system) based on a multi-channel (wideband) filter set. This system is much better than a narrow-band dry / step filter, wave filter or single filter. Frequency spectrometer (111〇11〇 ^]: 〇111 ^ 〇1 ^) system is much cheaper 'and our method greatly simplifies the hardware complexity of traditional spectral response measurement systems. (B) We have developed a QRCP

Page 25

= Method-based filtering selection algorithm, QRCp — based method, to efficiently design this MCF system. -In the present invention, the filter set selected by the QRCP-based method is more noise-sensitive than the MCF system constructed by other methods, that is, it has a higher accuracy for spectral response measurement. Fourth, in this embodiment of Ben Sheming, the spectral plasticization normalized root mean square error value (NRMSE) of all color channels obtained by the MCF system constructed by the filter set selected by the QRCP-based method is less than 〇22 ， And the maximum normalized estimation error value in these channels is 0.004 〇 2 5. The drawings and the description of the present invention have disclosed a preferred embodiment of the present invention ', although specific conditions are applied therein, but are only used to broadly explain the present invention' and are not intended to limit the present invention. The scope of the present invention is defined in section B of the patent application scope described later. 5 · References: (Patent Literature) [1] Y. Okui, s. Ito, and M. Sugiyama,

Multi-channel spectral light measuring device, US Patent No .; 4909633, issued date: March 20, 1 99 0.

Page 26 491938 V. Description of the invention (22) [2] B. Larsen, " Color camera system having complete spectral characteristics, ' US Patent No. 3789132, issued date: Jan 29, 1974.

[3] G. Abraham and G. Wenzel, " Method and apparatus for determining spectral sensitivity parameters of color-sensitive receptors in the eye, " US Patent No .: 5801808, issued date: Sept. 1, 1998.

[4] K. Takahashi and T. Terashita, ”Method of estimating spectral distribution of film and method of determining exposure amount, M US Patent No .: 5671060, issued date: Sept. 23, 1997.

[5] S. Osaki and M. Sugiyama, ’丨 Spectral sensitivity correcting device in a photoelectric tristimulus colorimeter, " US Patent No .: 4989982, issued date: Feb. 5, 1991.

[6] G. A. Zerlaut, R. D. Whitaker, and A. W.

Purnell, " Method and apparatus for determining spectral response and spectral response mismatch between photovoltaic devices, 丨 'US Patent No .: 4467438, issued date: Aug · 21, 1984 · [7] GW Chang and YC Chen, " Automatic Spectral Respons ivi ty Measurement System for Color Video Cameras, n (Automated Spectral Response of Digital Color Imaging Device

Page 27, 491938 V. Description of the invention (23) Measurement method and system), F i 1 edt 〇R 0 CP atent (N 〇 88109743, June 10, 1999) · (Technical data and papers) [8] G. Sharma, J. Trussell, `` Digital color imaging, IEEE Trans. Image Processing, vo 1. 6, no. 7, pp. 90 1-932, 1 997.

[9] JAC Yule, Principles of Co lor Reproduction, NY: John Wiley and Sons, 1967. (reprinted in 1989) [10] S. 0. Park et a 1., " Development of spectral sensitivity measurement system of image sensor devices, " Proc. of IS & T / SID 1995 Color Imaging C ο n f.: Col or Science, System and Appls ·, pp · 115-118, 1995.

[11] PM Hubei et a 1., " A comparison of methods of sensor spectral sensitivity estimation, M Proc. Of IS & T / SID 1994 Color Imaging Conf .: Color Science, System and Appls, pp. 45-48, 1994.

[12] G. Wyszecki, '丨 Multifilter method for determining relative spectral sensitivity functions of photoelectric detectors, M J. Opt.

Soc. Am., Vo1. 50, no · 10, pp · 992-998, I960 · [13] International Commission on Illumination

Page 28 491938 V. Description of the Invention (24) (CIE), Determination of the Spectral Respons i v i ty of Optical Radiation Detectors, Pubication CIE No. 64, Paris, France, 1982.

[14] G. W. Chang and Y. C. Chen, " Automatic spectral measurement system for col or video cameras, 丨 丨 Trans. Consumer Electronics, vo 1. 45, no · 1, 225-235, 1999.

[15] GW Chang and YC Chen, " Spectral Respons ivi ty Estimator for Color Vision Systems: Filter Selection and Noise Effect, 丨 丨 Proceedings of National Sc i. Council, Part A, vo1. 13, no. 1 (to appear in Mar. 2001).

[16] W. N. Sproson, Color Science in Television and Display Systems, Adam Hilger, Bristol, 1983.

[17] GE Healey and R. Kondepudy, " Radiometric CCD earnera calibration and noise estimation, " IEEE Trans. Pattern Anal. Mach. Intell ·, vo1. 16, no. 3, pp · 267-276, 1994 · [ 18] GH Golub and CF Van Loan, Matrix Computations, 2nd ed., Baltimore and London: John Hopkins Univ. Press, 1989.

[19] C. L. Lawson and R. J. Hanson, Solving Least Squares Problems, NJ: Prentice-Hall, 1974.

[20] V. Oppenhe i m and R. W. Schafer, Discrete-Time

Page 29 491938 V. Description of the invention (25)

Signal Processing, Prentice Hall, NJ, 1989.

[21] G. Wyszecki and W. S. Stiles, Color Science -Concepts and Methods, Quantitative Data and Formulae, John Wiley and Sons, NY, 1982.

[22] GW Chang, " Spectral Re spons ivi ty Estimation and Colorimetric Characterization for Machine Vision Systems, n (Machine Vision System Light, Spectral Response Estimation, and Color Mode Characterization) Ph.D. Dissertation, National Tsing Hua Univ ·, June 200 0 · (Unpublished, expected to be released in June 2002)

Page 30 491938 Brief description of the drawings Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 1 The spectral response characterization process in a multi-channel filtering (MCF) system. A preferred implementation example of the spectral radiation power distribution on the surface of the system light source. Effect of QRCP method on sub-set selection (su) 3Set selection. QRCP-based filter selection algorithm flowchart. The spectral penetration distribution of the filter bank selected by the QRCP-based method. Estimate the spectral response of the Sony XC711 camera. Normalized mean square error (NRMSE) distribution with respect to test color light (TCL) numbers i to 72. Based on the comparison of the estimated spectral response of the Λ 旒 noise ratio of 35 dB from (a) selection method} selection method 2, (C) selection method 3, and (d) QRCp_based selection method 'two. Parts to Table 1 Table 2 Table 3 Table 4 Table 5 The results of different filter selection (indicated by wave reduction). / Yes. A collection of tables compiled by Koukyo cited QRcp ~ based method performance evaluation table. = Number of conditions obtained from different filter selections. Verify the simulation results of noise effects. From the different filter selection methods, P is from the above > Miscalculated spectral response estimation

Claims (1)

491938 Six patents for patent application 1 ---------- κ A multi-channel design method for measuring the spectral response of a digital imaging device, in which the over-group of the multi-channel crossing wave system μ-wave system is specially selected and Spectral filtering installed in different channels 2 = Source and optical system 'The emitted light emitted by this light source is different in the filters that form the group in different channels, and produces different specifications: 间: y, 刀, 测: measured images Device; A method of selecting a wave setter for a mysterious multi-channel ferry wave system. The composition steps of the method are as follows: to index all the filters to be selected for the group, and index them to create a row vector form. Expressed (2) Collect the aggregate matrix of the spectral radiation distribution of the surface light source through all the stimuli to be colored; Generated by the selected spectral filter to form a huge color stimulus (3) Perform the singular value decomposition operation on the color stimulus matrix , Corresponding to the right singular matrix formed by all right singular vectors; (4) check whether the rank of the matrix of the color stimulus matrix is greater than or equal to the filter on the pass, and the condition of the total number of filters is true; if yes, proceed to the following One step; if not, it means that the appropriate filter set cannot be selected for this method; (5) Take the same number of rows in the right singular matrix as the previous total number of filters on the channel to construct the submatrix of the right singular matrix, called the right submatrix; 〃 (6) Take the The orthogonal triangle decomposition and row conversion operation of the transposed matrix 'and stores the permutation matrix generated by the operation; Page 32 491938 VI. Patent application scope (7) Multiply the column vector formed by the transposition of the filter index vector with the permutation matrix. The result is a transposed filter, and the waves are arranged in rows for the index vector. The column vector formed by the conversion; (8) Take the vector formed by the same number of elements in front of the column vector formed by the row permutation conversion as the total number of filters in the channel Λ, and call it the selected filter index Vector, all elements in the vector are the selected filter index number, so we have completed the filter selection work, and have completed the design of the optical system based on the multi-channel filter set. 2 · According to the design method of the multi-channel filtering system described in item 1 of the scope of the patent application, 'The selected set of spectral filters must be installed on the same movable disk device, respectively, and different spatial distributions are formed at the same time. Channel, so that the radiant light emitted by the common system light source penetrates different filters at different times to generate corresponding color stimuli and is incident on the image acquisition device to be measured according to item 1 of the patent application scope 3. ^. ., Not described 1 of the channel filter system Method, in which the number of selected wave receivers should be between 1 and 10. 4. According to the design method of the scope of the application for patents No. 1 or No. 2 and No. 10, where the multi-channel crossing wave described in the item No. should be greater than or equal to the number of filters selected In addition to the filter set, 1 θ is on the channel in addition to: Any other channel j must be installed Page 33 491938 VI. Patent application scope filter. 5. The design method of the multi-channel filtering system according to item 1 or 2 of the scope of the patent application, in which the number of channels on the multi-channel filtering system device should be between 1 and 200. Φ Bill p. 34